Another presidential election, another case of pollsters underestimating the Trump vote. While the magnitude of this year's miss was small, and in fact smaller than in 2016 and 2020, it still had a tell-tale sign of a miss – the misses were all in the same direction.

In 2016, polling averages underestimated the Trump vote in swing states by 2-4 points. In 2020 it was 3-5 points. In 2024 it was 2 points. Why do polls continue to underestimate the Trump vote?

The most likely explanation is that there is a small sector of the electorate that we and other polling organizations aren't able to properly reach, but also aren't able to identify, and thus don’t accurately account for in weightings. Though the miss is small, it is a blackmark for the the industry that we have not been able fully rectify the issue in eight years of trying.

You may have read of the “Trump Whale,” a Frenchman who bet $30m on a Trump win. Some prominent Twitter accounts are agog over the brilliance of the man, who, in his telling to a Wall Street Journal reporter, conducted a private poll where it was asked ‘Whom is your neighbor voting for’ as opposed to ‘Whom are you voting for,” and found support stronger in the latter and thus bet on a Trump win.

Asking "how does your neighbor feel about such and such" has long been a tactic pollsters use when trying to ascertain true public opinion, and was asked in many other polls by many other pollsters on this election.

We asked our own version of the question with "Who do you think will win the election?" That result was 46/41 for Trump, compared to 49/48 for Harris when it came to voting intention. Thus many people expected Trump support was higher than reported voting intention indicated, but unlike the Whale, most were not willing to bet $30mm on it being determinative.

You may be dubious of public opinion polling after three straight elections where it underestimated the Trump vote, but we humbly suggest you don’t throw the baby out with the bathwater. There are some nefarious pollsters who intentionally publish misleading results, or inept pollsters who do so unknowingly, but the majority of polls that make it into national media are credible pollsters that reliably provide useful information.

If a poll asked people what they thought of a detergent, and the result was 49%/48% for/against, and it turns out the “true” view of the entire population was 48%/49% the poll would still be of great value even though it missed by 2 pts — it told you people are very evenly divided on the detergent, which is true. Same if the poll were about a policy proposal that showed 63/37 for/against, when the "true" view was 60/40. The poll missed by three points but still told you what you needed to know — people are largely in favor of the proposal.

When it comes to an election the same is true, polls can reliably tell you who is preferred and by how much. What they can’t reliably tell you is who will win when the margins are razor thin.